Organizing public sector AI adoption: Navigating between separation and integration

Abstract

Artificial Intelligence (AI) has the potential to improve public governance, but the use of AI in public organizations remains limited. In this qualitative study, we explore how public organizations strategically manage the adoption of AI. Managing AI adoption in the public sector is complex because of the inherent tension between public organizations' identity, characterized by formal and rigid structures, and the demands of AI innovation that require experimentation and flexibility. Our findings show that public organizations navigate this tension either by creating separate departments for data science teams, or by integrating data science teams into already existing operational departments. The case studies reveal that separation improves the technical expertise and capabilities of the organization, whereas integration improves the alignment between AI and primary processes. The findings also show that both approaches are characterized by different AI adoption barriers. We empirically identify the processes and routines public organizations develop to overcome these barriers.